Oecologia (2001) 127:572–585
DOI 10.1007/s004420000620
Julian D. Olden · Donald A. Jackson
Pedro R. Peres-Neto
Spatial isolation and fish communities in drainage lakes
Received: 25 October 1999 / Accepted: 27 November 2000 / Published online: 22 February 2001
© Springer-Verlag 2001
Abstract Fifty-two drainage lakes, located in south-central Ontario, Canada, were examined to study the association of isolation- and environment-related factors with
fish community composition. Eight quantitative measures of lake isolation were examined, each of which incorporated potential ecological “challenges” that a fish
encounters when moving between lakes. A Procrustean
approach was employed to assess the degree of concordance between fish assemblage structure, measures of
lake isolation and environmental conditions (i.e., lake
morphology and water chemistry). Our results revealed a
high concordance between patterns in fish community
composition and lake isolation and lake morphology at
the watershed scale, suggesting that insular and habitatrelated factors influence the structure of fish communities. At the scale of the individual lake, this relationship
varied greatly, ranging from a strong match of community composition with both spatial and abiotic conditions
to communities exhibiting weak association with these
conditions. Furthermore, we showed that alternative
measures of lake isolation provide additional insight into
potential factors shaping patterns in fish community
composition; information not provided using straightline distances between lakes. Finally, the statistical methodology outlined in this paper provides a robust technique for assessing both the overall association between
multivariate data matrices (i.e., landscape or regional
scale), as well as facilitating the examination of smallerscale relationships of individual observations (i.e., local
scale).
Keywords Extinction · Colonization · Dispersal · Lake
connectivity · Isolation
J.D. Olden (✉) · D.A. Jackson · P.R. Peres-Neto
Department of Zoology, University of Toronto, 25 Harbord Street,
Toronto, Ontario, M5S 3G5, Canada
Present address:
J.D. Olden, Department of Biology, Colorado State University,
Fort Collins, CO 80523-1878, USA
e-mail: olden@lamar.colostate.edu
Tel.: +1-970-4912414, Fax: +1-970-4910649
Introduction
Factors regulating the species composition of disjunct
habitats (e.g., islands) have been a major focus of ecological and evolutionary studies since the work of Darwin and Wallace. Most studies have been directed at how
abiotic environmental conditions or biotic interactions
contribute to the maintenance of populations. MacArthur
and Wilson (1967) focused attention on the role of colonization and extinction on islands, and these ideas were
extended to other insular habitats (e.g., Brown 1971;
Barbour and Brown 1974). Levins (1969) and Hanski
(1982, 1991) showed that the movement of individuals
between patches or insular habitats contributes to survival and distribution of a species. Individual species differ
in their probability of colonizing patchy habitats due to
differential abilities to disperse and the effects of varying
distances between patches. As a result, sites exhibiting
similar environmental characteristics may have dissimilar faunas because of differential chances of colonization
and extinction for each species. Terrestrial ecologists
have found these ideas useful for understanding systems
containing highly fragmented habitats (Harrison et al.
1988; Paine 1988; Fahrig and Merriam 1994). Several
studies have considered how readily individuals may
move between patches and the longevity of populations
in patches with various degrees of connectivity. It is believed that more isolated patches will have reduced rates
of colonization, equal or more extreme extinction rates
(reduced “rescue effect”; Ray and Gilpin 1991) and, as a
consequence, a more depauperate community. Changes
in the degree of isolation through alterations to patch
connectivity may lead to changes in population viability
and community composition. More recently, ecologists
have applied these ideas of patch or metapopulation dynamics to issues of conservation [see Saunders et al.
(1991) for a review].
Despite the substantial amount of empirical and theoretical work on patch dynamics and the role of habitat
connectivity in terrestrial ecosystems, there has been
limited application to aquatic ecosystems (see Jackson et
573
al. 2001. This lack of attention is surprising given that
many aquatic ecosystems provide direct counterparts to
patch habitats (e.g., stream sites and lakes) and corridors
(i.e., streams, rivers). Furthermore, compared to terrestrial studies where researchers must often assume the dispersal pathway between habitat patches, aquatic organisms are confined to the interconnecting streams and rivers between water bodies. Therefore, lakes within a watershed provide a unique study system where inter-patch
distances can be more readily defined and quantified.
The patchy, insular nature of seepage and drainage lakes
influences both the colonization and persistence of local
fish populations, thereby influencing fish community
composition (Tonn and Magnuson 1982; Tonn et al.
1990; Magnuson et al. 1998; Matthews and Robison
1998). Fish populations in lakes that are more isolated
from other waterbodies will be more susceptible to local
extinction and more limited in their recolonization
rates. On the other hand, high rates of fish dispersal between lakes can result in great similarities in fish community composition (homogenization sensu Radomski
and Goeman 1995) among connected lakes. Although
seepage lakes are completely isolated from one another
(except during periods of extreme flooding or due to
stocking activities by humans), the degree of isolation
among drainage lakes is determined primarily by the
length of adjoining watercourses, and by the type and
number of natural and artificial obstacles. Natural obstacles or barriers include temporary factors such as beaver
dams, seasonal drying or warming of waterways, as well
as permanent channel characteristics such as strong currents, riffles, cascades and waterfalls. In addition, rivers
that contain a strong piscivore may have reduced movements of fishes among tributaries due to predation on potential dispersers (Townsend and Crowl 1991; Fraser et
al. 1999), or may promote movement by inducing dispersal of individuals from side pools and channels
(Fraser et al. 1999).
The effects of spatial isolation among habitat patches
in streams, rivers and lakes may have considerable implications for understanding the main forces shaping fish
population and community dynamics, and for the management and conservation of our aquatic resources. Relationships between habitat conditions and population
characteristics (e.g., spatial distribution, abundance)
could be confounded if lake isolation causes a reduction
in the dispersal rates of fish species, thereby excluding
them from potentially available habitat. In such cases,
isolation-related factors could decouple associations between fish community composition and the surrounding
abiotic environment (e.g., Angermeier and Schlosser
1989; Snodgrass et al. 1996; Snodgrass and Meffe 1998).
Ultimately, this could affect our understanding of population and community ecology of fishes, influence our
interpretation of factors controlling lake food webs (e.g.,
Hershey et al. 1999), and reduce or even eliminate the
validity of biological assessment measures [e.g., “index
of biotic integrity” (Karr 1981)]. Knowledge regarding
the spatial connectedness of lakes is essential for under-
standing and predicting the potential dispersal and colonization of introduced species within any watershed. For
example, introductions of strong littoral predators (e.g.,
smallmouth bass, northern pike) likely result in the local
extinctions of many small-bodied species (Jackson and
Harvey 1989; Robinson and Tonn 1989; Jackson et al.
1992; Chapleau et al. 1997; Findlay et al. 2000; MacRae
and Jackson 2001). Therefore, the fish communities of
many lakes may be affected by the introduction of a single predator into a single lake via dispersal and subsequent biotic interactions with native fishes. As a result,
the management and stocking of lakes must be considered in a watershed context incorporating interconnections among lakes, rather than on a lake by lake basis as
decisions made regarding one system may impact significantly on neighbouring lakes within a short time frame.
Although various field and modelling approaches have
been used in studying habitat isolation in terrestrial ecosystems, there are few examples involving aquatic studies. Researchers have often developed relatively simple
measures of lake isolation, including straight-line distance
(e.g., Magnan et al. 1994), watershed area (e.g., Tonn
and Magnuson 1982), lake elevation (e.g., Magnuson
et al. 1998), presence or absence of intermittent or permanent inlets and outlets (e.g., Rahel 1986; Robinson
and Tonn 1989), and watercourse distance to the nearest
neighbouring lake (e.g., Tonn et al. 1990). In many cases
it is implied that colonization of a lake only originates
from the closest, adjacent lake. However, fish could
move more widely over time within a watershed, for example using a series of lakes as “stepping stones” for
colonization (Magnuson 1976). Clearly, there is a need
for a more formal examination of the role of lake isolation in shaping fish communities. More quantitative
measures of spatial isolation, incorporating different
characteristics of the watershed that contribute to the insular nature of lakes, are required. Gaining this knowledge will aid our understanding of the role that isolation
plays in determining species diversity and community
composition. In the present paper, we have the following
objectives:
1. Examine several published and newly proposed quantitative measures of spatial isolation for drainage
lakes, each of which provide large-scale or global
measures of the challenges fish encounter when dispersing among lakes.
2. Assess the relationship between patterns in fish community composition and factors related to the degree
of lake isolation and environmental conditions, at
both the lake and watershed scale.
3. Demonstrate the use of Procrustean methods for examining these relationships between the data sets, and
highlight their potential use for assessing both the
overall degree of concordance between data sets and
facilitating the examination of smaller-scale relationships among individual observations.
574
Fig. 1 Schematic map of the
study lakes located in the Black
and Hollow River watersheds;
inset illustrates inter-lake distances between Jill Lake and
Ridout Lake for example calculations in Materials and methods
Materials and methods
Ecological data
This study focuses on 52 drainage lakes located within the Black
and Hollow River watersheds of south-central Ontario, Canada
(Fig. 1). The lakes are located within a heterogeneous landscape
with lake elevation ranging from 320 m to 488 m above sea level.
Most lakes contain one or more inlets and a single outlet, and are
connected to one another by watercourses ranging in distance
from 63 m to 138 km. Each lake was surveyed to determine fish
community composition, as well as quantifying lake morphological and water chemistry characteristics. Intensive sampling using
fine- and coarse-meshed trap nets, baited minnow traps, seine
nets, clear plastic traps, and multi-meshed gill nets (see Jackson
and Harvey 1989, 1997), was conducted to estimate the presence/absence of each fish species. Sampling was conducted continuously for 3–7 days and nights using all types of sampling gear
in all available habitats to maximize the number of species captured (see Jackson and Harvey 1989; Jackson et al. 1992 for details). Increased effort was used in larger lakes and some lakes
were sampled repeatedly at different time periods to determine
whether additional species could be captured. No additional species were found during these re-sampling periods. Species not
captured after intensive sampling were assumed to be absent or so
rare as to be of minimal importance for community comparisons.
From the 31 fish species captured, those found in 5% or more of
the study lakes were used for the analyses (Table 1). A series of
abiotic variables were also measured for each lake (Table 2). Lake
morphological variables included lake area, volume, maximum
depth, shoreline perimeter, and island perimeter. The water chemistry dataset consisted of pH, alkalinity, conductivity, colour, and
concentrations of calcium, sodium, chloride, dissolved organic
carbon, dissolved inorganic carbon, magnesium, potassium and
sulphate [see Jackson (1988) for details].
Measures of lake isolation
This section describes the calculation of eight lake isolation measures. In all cases, distances were calculated from 1:50,000 topographic maps, and were recorded in kilometers. After each isolation measure is described, we provide an example of the calculation using Ridout Lake (lake no. 18) and Jill Lake (lake no. 21), illustrated in the inset of Fig. 1.
The latitudinal and longitudinal position of each lake was used
to generate an inter-lake distance matrix based on SL-dis, i.e.,
Euclidean distance, between the center of each of the lakes. Given
the limited latitudinal variation involved, no correction was made
575
Table 1 List of fish species, including species abbreviation
(Code) and occurrence (%) in
the 52 study lakes
Table 2 Distribution parameters [median, lower quartile
(LQ), upper quartile (UQ),
minimum (Min.) and maximum
(Max.)] for lake isolation, lake
morphology and water chemistry variables
Common name
Scientific name
Code
Occurrence
Blacknose dace
Blacknose shiner
Bluntnose minnow
Brook stickleback
Brook trout
Brown bullhead
Burbot
Cisco
Creek chub
Common shiner
Fathead minnow
Finescale dace
Golden shiner
Iowa darter
Lake chub
Lake trout
Largemouth bass
Northern redbelly dace
Pearl dace
Pumpkinseed
Rock bass
Smallmouth bass
White sucker
Yellow perch
Rhinichthys atratulus
Notropis heterolepis
Pimephales notatus
Culaea inconstans
Salvelinus fontinalis
Ameiurus nebulosus
Lota lota
Coregonus artedi
Semotilus atromaculatus
Luxilus cornutus
Pimephales promelas
Phoxinus neogaeus
Notemigonus crysoleucas
Etheostoma exile
Couesius plumbeus
Salvelinus namaycush
Micropterus salmoides
Phoxinus eos
Margariscus margarita
Lepomis gibbosus
Ambloplites rupestris
Micropterus dolomieu
Catostomus commersoni
Perca flavescens
BND
BNS
BNM
BS
BT
BB
B
C
CC
CS
FHM
FSD
GS
ID
LC
LT
LMB
NRD
PD
PKS
RB
SMB
WS
YP
10
21
31
12
35
67
17
6
75
31
17
23
44
8
6
15
21
44
25
88
6
31
79
85
Median
LQ
UQ
Min.
Lake isolation
Watercourse distance (km)
Elevation (m)
30.8
350
19.3
335
51.4
387
0.06
320
138.1
488
Lake morphology
Area (ha)
Maximum depth (m)
Shoreline perimeter (km)
Volume (×104 m3)
35.7
13.4
3.8
23.3
16.8
8.1
2.5
6.8
62.8
23.7
6.0
52.0
5.5
3.0
1
1.9
2821.0
73.2
104.7
6150.0
Water chemistry
Alkalinity (Eq L–1)
Calcium (mg L–1)
Chloride (mg L–1)
Colour (platinum units)
Conductivity (mhos/cm)
Dissolved organic carbon (mg L–1)
Dissolved inorganic carbon (mg L–1)
Magnesium (mg L–1)
pH
Potassium (mg L–1)
Sodium (mg L–1)
Sulphate (mg L–1)
27.5
2.58
0.30
13.50
29.00
3.95
0.80
0.69
6.00
0.40
0.69
6.17
15.5
2.35
0.20
9.00
24.88
3.30
0.40
0.60
5.74
0.34
0.54
5.51
48.5
2.87
0.40
22.88
31.00
5.20
1.10
0.84
6.40
0.45
0.85
6.89
0.00
2.05
0.05
1.00
9.00
1.20
0.20
0.43
4.79
0.21
0.41
2.69
to account for convergence of longitudinal lines. Commonly, the
straight-line distance (SL-dis) between sampling sites is used as a
measure describing the spatial arrangement of habitats within the
landscape (e.g., Magnan et al. 1994; Sjögren Gulve 1994; Little et
al. 1997). For this reason, we chose to use this spatial measure.
The SL-dis between Ridout Lake and Jill Lake is 2.5 km.
Watercourse distance (WC-dis) between each study lake was
calculated as the shortest distance between two lakes following the
connecting waterways. The WC-dis between Ridout Lake and Jill
Lake is 7.8 km. In addition, the location of lakes along the watercourse can be considered as a series of stepping stones (sensu
MacArthur and Wilson 1967), such that fish could colonize lakes
in steps, thus aiding the movement of individuals between distant
Max.
122.0
3.68
7.50
230.0
49.50
18.0
2.00
1.97
6.90
5.62
4.60
34.00
lakes. Arguably, two lakes in a chain with a third lake between
them may be less isolated than two lakes lacking such an intermediate lake (Magnuson 1976). For this reason, the WC-dis excluding the distance traveled through other intermediate lakes might
more accurately represent the distance between a set of lakes
(WC-body). The WC-body connecting Ridout Lake and Jill Lake
is 4.2 km (equal to 7.8 km minus 3.6 km).
For fish moving between lakes, movement downstream may
represent a more passive dispersal, whereas movement upstream
requires energetic expenditures. To incorporate the fact that upstream lakes are potentially more difficult to colonize, we developed two measures that represent the distance between lakes that
involves only upstream movement. Since the upstream distance
576
from lake A to B is not necessarily equal to the upstream distance
from lake B to A, an asymmetric distance matrix was constructed.
This matrix was subsequently decomposed into two components.
The first matrix was the symmetrical component (US-sym), which
was the average of the upstream distances between two lakes (i.e.,
the average of A to B and B to A). The upstream distance from
Ridout Lake to Jill Lake is 1.9 km, and the upstream distance from
Jill Lake to Ridout Lake is 4.5 km, resulting in an US-sym of
3.2 km (average of 4.5 km and 1.9 km). The second matrix was
the skew component (US-skew), which was calculated as the difference between the average upstream distance between these
lakes (US-sym) and the upstream distance between lake A and B.
The US-skew between Jill Lake and Ridout Lake equals 1.3 km,
which is calculated as 4.5 km (US-sym) minus 3.2 km (the upstream distance from Jill Lake to Ridout Lake). Similarly, the
US-skew between Ridout Lake and Jill Lake is –1.3 km (equal to
1.9 km minus 3.2 km). The magnitude of US-skew is the same between the lakes (1.3 km), therefore, only the positive US-skew
value was retained. The decomposition of the original asymmetric
upstream distance matrix into two components was required to ensure that we accounted for all of the information regarding the upstream distance between the lakes. The symmetric portion
(US-sym) represents the average upstream distance between lakes
and the US-skew provides a measure of the distance between a
pair of lakes in terms of their relative upstream components. Both
US-sym and US-skew describe the potential difficulties a fish faces when traveling upstream, such as the general constriction of the
watercourse closer to headwaters, greater probability of encountering obstacles (i.e., riffle areas or waterfalls), and traveling
against the flow of water.
To explore patterns of lake connectivity within the watershed,
we constructed a hydrographic tree (Magnan et al. 1994). At each
bifurcation in the tree, a node number was assigned. A lake-bynode matrix was constructed by indicating each node (assigning a
value of 1) that connects the lake to the origin or “root” of the
tree, and 0 to all the remaining nodes. This procedure results in
each lake being classified by a row vector of binary values (0 or
1). An inter-lake Euclidean distance matrix (hydrographic tree;
HG-tree) was constructed based on the lake-by-node matrix. Hydrographic trees provide a quantitative measure of the number of
different streams changes that a fish encounters when dispersing
from one lake to another and therefore, represent increasing isolation between any two lakes [see Magnan et al. (1994) for more details]. Although this approach provides a relative measure of the
order of differences between pairs of lakes, it loses all information
related to the magnitude of the scale of distances between lakes.
The last two measures of spatial isolation incorporated WC-dis
and difference in elevation between the study lakes. Two distance
matrices were constructed where the first represented the WC-dis
and the second was based on the difference in elevation between
lakes. Both of these 52-by-52 distance matrices (i.e., 52 lakes)
were used to produce two isolation measures in the following
manner. Both distance matrices were unfolded from a triangular
form into a continuous vector [see Legendre and Legendre (1998),
p 560]. A matrix of size 1,378 by 2 (i.e., the (52×51)/2+52 rows
and two columns) was created by combining the two vectors. A
principal component analysis (PCA) (the equivalent of a model II
regression) was employed using these two variables. The first axis
summarized the general allometric relationship between inter-lake
WC-dis and elevation difference. This set of 1,378 observations
from the first axis was then converted back into a triangular matrix
of inter-lake distances. This matrix (D-elev1) provides a generalized measure of the overall distance, both by water and elevation,
between any two lakes. The second axis (i.e., PCA axis II) provides a measure uncorrelated with the information on axis I and
represents the relative change in elevation compared to distance
by water (i.e., stream gradient measure). In a similar fashion, the
1,378 observations from axis II were converted back into a triangular distance matrix (D-elev2). These two measures of spatial
isolation incorporated the notion that fish may have greater difficulty gaining access to higher elevation lakes within a watershed
because of the energetic costs associated with swimming against
the water current and increased probability of encountering barriers.
In summary, each of the measures of lake isolation described
above were represented using lake-by-lake distance matrices that
described the degree of lake isolation based on ecological considerations of dispersal. Each of these measures quantified potential
“challenges” a fish may encounter when moving between lakes,
incorporating the idea that the characteristics of the interconnecting streams and rivers could contribute to the insular nature of
many lakes.
Statistical analyses
To summarize compositional patterns in fish community assembly,
correspondence analysis (CA) was conducted on the species presence/absence data matrix. CA was chosen because it is not affected as adversely by data that contains numerous zero values (Gauch
1982) as are many other ordination methods. To summarize patterns in environmental conditions among the lakes, PCA of the
correlation matrices was employed to ordinate the lakes separately
on the basis of water chemistry and lake morphology. All environmental variables (except pH) were log (x+1) transformed prior to
PCA to linearize bivariate relationships. Principal coordinate analysis (PCoA) was performed on each isolation-distance matrix to
reduce the number of dimensions and summarize the majority of
the variation in the original data into a subset of axes. PCoA was
used because of its ability to analyse a wide range of distance (association) matrices, which was essential due to the nature of our
isolation measures. To facilitate interpretation of the results in all
cases only the first two dimensions were retained since they explained the majority of the variation in the original data.
Concordance between lake ordination scores from the two-dimensional multivariate summaries of the fish communities, isolation measures, lake morphology, and lake water chemistry were
compared using a Procrustean approach. Procrustes analysis is a
superimposition method which in the simplest case, compares a
pair of ordinations by using a rotational-fit algorithm that finds the
optimal match between corresponding observations of the ordinations (see Gauch 1982; Digby and Kempton 1987; Jackson 1995;
Legendre and Legendre 1998). Figure 2 illustrates the steps of
Procrustes analysis, using two configurations each containing
three data points (e.g., three lakes: A, B, and C). The most commonly used Procrustean fitting method is based on the least-squares criterion, which superimposes the coordinates of observations from
one ordination onto those of another ordination so as to minimize
the total squared distances between corresponding points (m2 statistic: Gower 1975; Rohlf 1990; Rohlf and Slice 1990). A disadvantage of the least-squares approach is that the superimposition
of two ordinations can often result in a general lack of fit for most
points when one or more atypical observations are present. A single atypical data point can exert a great influence due to the leastsquares criterion resulting in all other points being “pulled” away
from their corresponding data points in the other ordination. This
effect tends to average the error across all points rather than the
single atypical one, potentially masking the true underlying relationship. The result can be to make atypical points appear acceptable or represent typical observations as potential outliers, and as a
consequence make the direct interpretation of residuals misleading
(Siegel and Benson 1982; Rohlf and Slice 1990). Such situations
are well recognized in methods such as regression analysis (Chen
et al. 1994).
To account for the potential shortcomings of the least-squares
method, Siegel and Benson (1982) described a resistant-fit approach for optimal superimposition of data configurations or ordinations in Procrustes analysis. Resistant-fit techniques are similar
to the nonparametric analogs of least-squares regression analysis,
and minimize the potentially strong influence of unusual observations. With this fitting technique, the influence of a few atypical
points is negligible and the interpretation of residuals is appropriate. A number of resistant-fit algorithms exist for Procrustes analysis and in this analysis we used the repeated-medians algorithm.
577
Fig. 2A–D Operations involved in comparing two configurations using Procrustes
analysis. A Original configurations or ordinations where the
apices of the triangles are analogous to lake positions in an
ordination, B ordinations after
translation and standardization
(configurations share a common centroid and scale),
C ordinations after reflection,
D ordinations after rotation
and dilation. Lower case letters
represent the rotated ordination
(i.e., fish community), and upper case letters represent the
target ordination (i.e., lake isolation, lake morphology or water chemistry). Reflection, rotation and dilation of the rotated
ordination are performed to
minimize the sum of the
squared distance (m2) between
corresponding observations
(i.e., lakes) of the two ordinations
Other robust and resistant methods might be used, but these methods generally have lower breakdown values than repeated medians
(Siegel and Benson 1982). The breakdown value describes the proportion of the data that can be changed without greatly influencing
the estimated value, i.e., a measure having a low breakdown point
may be affected even if only a few atypical points are present (e.g.,
Chen et al. 1994). Siegel (1982) showed that the repeated median
algorithm has a breakdown point at nearly 50%, which is higher
than least-squares (0%) or alternative techniques such as the single
median method (29%). Therefore we employed resistant-fit Procrustes analysis using the repeated medians algorithm.
To statistically test the degree of concordance between fish
community composition, lake isolation measures, lake morphology, and lake water chemistry, a Procrustean randomization test
[PROTEST: Jackson and Harvey (1993); Jackson (1995); Legendre
and Legendre (1998)] was used. PROTEST estimates the significance of an observed m2 statistic with the underlying null hypothesis being a random association between the two datasets (e.g., fish
community and lake morphology ordinations). This procedure randomly permutes the original observations of one matrix in a way
that each lake can be assigned to any of the possible values attributed to other lakes (Jackson 1995). This results in one matrix being unchanged, and the second matrix being randomly permuted,
but maintaining its within-matrix covariance structure. After each
permutation, the m2 statistic between the two matrices is recalculated. The proportion of calculated m2 statistics, including the observed m2, that is smaller than or equal to the observed m2 statistic
provides the significance level of the test. In this study, 9,999 random permutations of the data were employed in order to ensure
the stability of the probability estimates (Jackson and Somers
1989). In each comparison, the lake ordination scores defined by
fish community composition was the rotated ordination (e.g.,
Fig. 2a–c), whereas the lake morphology, water chemistry and
lake isolation ordinations were considered the target and therefore
were fixed (e.g., Fig. 2A–C). The analysis was performed in this
manner because the isolation, morphology, and chemistry matrices
were considered to be the independent variables, and the species
composition to be dependent on the degree of connectivity with
other lakes in the watershed and habitat conditions. Procrustes
analysis was conducted using the GRF-ND software (Slice 1994).
Although alternative approaches, such as the Mantel test or canonical correspondence analysis (CCA), allow us to compare two
or more matrices, the Procrustean approach has advantages over
them given our specific goals. First, the Mantel test is simply a
correlation between distance matrices and provides no insight into
which observations show a good match between their fish community composition relative to lake isolation and environment, and
which ones do not. Similarly, CCA tests for overall relationships
between data sets, but does not show the degree of concordance
for each observation. Given that we are interested in determining
how closely each lake matches in its community composition and
isolation/environment conditions, the Procrustean method is ideal.
In addition, we wish to examine whether principal patterns of variation in the community match with principal patterns in lake isolation measures and environmental conditions, which is not assessed
using CCA.
Results
CA of the species presence/absence dataset provided a
two-dimensional summary of dominant patterns in fish
assemblage structure among the study lakes. The results
show that as you progress positively from left to right
along axis I and from top to bottom along axis II
(Fig. 3A), the species composition shifts from communities containing primarily small-bodied fishes (mainly cyprinids) to communities containing large-bodied fishes
(e.g., Micropterus spp.). Lakes located in the top-left
corner of Fig. 3B contain relatively greater numbers of
smaller-bodied species compared to lakes located at the
bottom and right of Fig. 3B which contain proportionally
greater numbers of large-bodied species. Furthermore,
lakes clustered close together in Fig. 3B share similar
fish community compositions, whereas lakes positioned
578
Fig. 3 Association of A fish
and B lakes based on correspondence analysis of fish species presence-absence. Letters
in A refer to species codes in
Table 1, and numbers in B refer
to lakes depicted in Fig. 1
at opposite ends of the plot contain distinctly different
fauna.
At the scale of the watershed (i.e., the complete set of
study lakes), Procrustes analysis assesses the overall degree of concordance between patterns in community
composition, lake isolation and habitat-related conditions. The results from the Procrustes analyses indicated
that the fish community ordination was significantly associated with SL-dis (m2=0.752, P=0.001), HG-tree
(m2=0.810, P=0.001), WC-dis (m2=0.780, P=0.012),
WC-body (m2=0.843, P=0.007), D-elev1 (m2=0.736,
P=0.001), US-sym (m2=0.772, P=0.010), US-skew
(m2=0.776, P=0.028) and marginally non-significant
with D-elev2 (m2=0.884, P=0.068). The fish community
ordination was also significantly correlated with the ordination of the lakes based on morphological characteristics (m2=0.839, P=0.001), but not with water chemistry
(m2=0.971, P=0.998). To ensure that relationships between these ordinations were not confounded due to inter-correlations between the lake isolation and morpholo-
gy data sets we compared these ordinations and found
that lake morphology and all measures of isolation were
not significantly concordant (m2=0.937 to 0.969,
P=0.209 to 0.980). This shows that the morphological
characteristics of the lakes were not spatially autocorrelated and suggests that isolation-related and lake morphology factors are contributing independently to the observed patterns in fish assemblages found in these lakes.
In addition to assessing overall relationships between
data sets, Procrustes analysis also facilitates a more detailed examination of the superimposition of the ordinations upon each other. That is, similarities and differences in the position of each of the individual lakes between the ordinations can be examined. Deviation in the
position of a particular lake on two superimposed ordinations is represented by a line segment called a vector residual. The length of the vector residual represents the
magnitude of the residual or lack of fit. Studying these
deviations provides a useful diagnostic tool because a
small vector residual indicates a close match and a large
579
Table 3 Ranked vector residuals for a subset of lakes from the
comparison of fish community ordination and ordinations based
on measures of lake isolation (SL-dis, HG-tree, WC-dis, WC-body,
D-elev1, D-elev2, US-sym, US-skew), lake morphology (Morph)
and water chemistry (Chem). Ranks range from 1 for the largest
vector residual, indicating a weak match between the position of
Lake
McDonald L. (12)
South McDonald L. (13)
Big Orillia L. (14)
Little Orillia L. (15)
Teapot L. (16)
Poker L. (5)
Wrist L. (8)
Big East L. (3)
Cinder L. (9)
Jill L. (21)
Herb L. (33)
Poorhouse L. (44)
Little Wren L. (30)
Ernest L. (32)
Sunken L. (34)
Kawagama L. (35)
Crosson L. (7)
Red Chalk L. (22)
the lake defined by the two ordinations, to 52 for the smallest vector residual, indicating a strong match between the position of the
lake defined by the two ordinations. Numbers in parentheses represent study lakes depicted in Fig. 1. Lakes are grouped together
in the table according to shared patterns in vector residual ranks
for all ordination comparisons.
Ranked vector residual
SL-dis
HG-tree
WC-dis
WC-body D-elev1
D-elev2
US-sym
US-skew Morph
Chem
15
9
13
17
8
5
1
3
14
29
44
35
22
42
49
20
2
6
10
8
14
14
17
3
2
1
11
23
32
44
51
31
29
19
9
7
7
4
8
11
9
6
10
3
14
22
47
41
28
52
40
17
13
12
8
6
19
20
18
7
4
3
9
35
31
48
39
32
26
15
5
11
7
5
3
4
9
10
11
12
21
17
49
50
33
52
51
18
16
8
8
4
9
12
10
6
7
3
15
21
46
41
28
52
37
16
11
5
12
10
13
11
4
3
8
1
7
20
49
43
25
51
36
16
5
9
1
5
6
3
4
2
12
24
45
8
11
7
47
20
39
40
42
48
Fig. 4 Standardized vector residuals from the comparison of fish
community composition with lake morphology (Morph) and water
chemistry (Chem). Numbers above bars represent study lakes depicted in Fig. 1
24
17
23
25
12
7
14
4
16
35
33
15
49
36
32
39
19
5
5
14
1
11
13
49
52
8
17
46
47
35
2
6
3
9
33
42
vector residual indicates weak association for a particular lake between the two ordinations being compared. In
this case, the vector residuals are represented by a line
joining the best-fit position of a given lake (observed
fish community composition) and the target position of
that lake (lake isolation, lake morphology or water
chemistry ordination). Therefore, the magnitude of the
vector residual reveals which lakes contain fish assemblages that agree with expectations based on lake isolation or habitat conditions (small vector residual) and
which lakes deviate from these expectations (large vector residual). For clarity we do not present the vector residual plots illustrating the superimposition of each pair
of ordinations, but rather present the standardized values
of the vector residuals for each lake (Figs. 4 and 5). Residuals were standardized to range between 0 and 1 to
make the results scale independent, and lakes were ordered (in decreasing order) based on their standardized
vector residual to facilitate comparisons among results.
Ranks ranged from 1 to 52 (i.e., number of lakes) where
a rank of 1 indicates the lake having the poorest match
between fish community composition and isolation/environment conditions and a rank of 52 indicates the strongest match compared to other lakes in the watershed.
Figs. 4 and 5 show the standardized vector residuals for
lake morphology and water chemistry, and lake isolation,
respectively. The vector residuals show that the degree
of fit between community, environment and isolation ordinations varied greatly across lakes, despite the fact that
community composition was highly correlated to patterns in lake isolation and morphology, and uncorrelated
with water chemistry at the watershed scale.
To demonstrate the interpretation of Procrustean results, we present a set of lakes that exhibit similar trends
580
581
in their ranked residual vectors. Table 3 shows that the
fish communities of some lakes exhibited a strong association with some measures of lake isolation (lakes
12–16: hereafter referring to the numbering of the lakes
on Fig. 1), some with lake morphology (lakes 5,8), some
with chemistry (lakes 3,9), some with lake morphology
and isolation (lakes 21, 33, 44), others with water chemistry and isolation (lakes 30, 32, 34, 35), and still others
match with lake morphology and chemistry (lakes 7, 22).
There were also a number of differences among the isolation measures. For example, Big Orillia Lake (lake 14)
and Little Orillia Lake (lake 15) demonstrated a strong
association with WC-body and D-elev1 and Teapot Lake
(lake 16) exhibited good concordance with HG-tree and
WC-body (Table 3, Fig. 5), showing that these measures
were more successful in explaining patterns in species
composition in these particular lakes compared to the
other isolation measures. A number of lakes (e.g., lakes
3, 12–15) exhibited contrasting fits based on D-elev1
compared to D-elev2 (Table 3). This was interesting
since both of these measures incorporate differences in
WC-dis and elevation among the lakes, but each describe
different aspects of this relationship. Finally, a number of
lakes showed poor agreement with SL-dis, whereas they
showed close agreement with other measures of lake isolation (e.g., lakes 8,13,16,23,28: Fig. 5).
Discussion
Watershed patterns in fish communities, lake isolation
and environmental factors
▲
Communities are rarely structured by a single factor;
rather, they are usually determined by an array of factors
acting simultaneously. In temperate North America, variables describing biotic, abiotic and lake isolation conditions have been frequently identified as important factors
shaping fish species richness and community composition (e.g., Johnson et al. 1977; Harvey 1975, 1978, 1981;
Tonn and Magnuson 1982; Eadie et al. 1986; Jackson
and Harvey 1989, 1993; Jackson et al. 1992; Matthews
and Robison 1998). The results from our study indicate a
strong association between fish community composition
and factors related to lake morphology, but not water
chemistry, for all study lakes. This suggests that wholelake attributes of lakes are important determinants of fish
community composition. Lake morphological factors include surface area, volume and shoreline and island perimeter which are correlated with habitat diversity (Eadie
and Keast 1984), and maximum depth which is negatively correlated with winter dissolved-oxygen concentraFig. 5 Standardized vector residuals from the comparison of ordinations based on fish community composition with lake isolation
ordinations. Numbers above bars represent study lakes depicted in
Fig. 1. SL-dis Straight-line distances; HG-tree hydrographic tree;
WC-dis, WC-body, D-elev1, D-elev2, US-sym and US-skew see
Methods and materials for description of isolation measures
tions and related to thermal stratification (Tonn and
Magnuson 1982; Jackson and Harvey 1989).
In addition to lake morphology, the results suggest
that the spatial configuration and degree of interconnectedness of the lakes within the watershed play important
roles in structuring patterns in community structure.
Similar to the role of the environment, isolation-related
factors have also been shown to be important determinants shaping temperate fish assemblages (Barbour and
Brown 1974; Magnuson 1976; Eadie et al. 1986; Jackson
1988; Tonn et al. 1990; Magnuson et al. 1998). The fact
that alternative measures of lake isolation were highly
concordant with patterns in fish community composition
emphasizes the fact that developing and testing isolation
measures other than straight-line distance is an important
exercise and can provide important additional insight into the role of isolation in shaping fish communities. Alternative measures of lake isolation incorporated the idea
that fish movement between lakes (and within/between
rivers) occurs predominantly via corridors of aquatic
habitat, and therefore colonization rates among lakes
will depend primarily on the number, length and suitability of these corridors. Watercourses with unsuitable characteristics could limit, or even preclude, the movement
of fish from one lake to another. Factors might include
the number of stream changes or confluences, and
stream-habitat characteristics of the connecting waterways (e.g., channel width, depth), which vary from pools
to swift-flowing riffles to waterfalls and dams. These
conditions may represent an increasing series of obstacles to fish moving between lakes, but present a greater
obstacle for species moving upstream compared to
downstream. Therefore, assuming a suitable habitat is
present, species could colonize more readily downstream
and species richness should be greater, on average,
downstream. Depending on each species’ size, swimming speed, and thermal tolerances, such stream conditions may represent species-selective filters or barriers.
Small-bodied and poor-swimming species may be incapable of moving upstream past riffles, minor cascades or
waterfalls, whereas larger and fast-swimming species
may pass with varying degrees of success. Cool- and
cold-water species may be restricted in their movement
to colder periods of the year or prevented entirely due to
low water levels and ice during the winter. As a consequence, upstream colonizing species may pass through
species-selective “filters”, which, in combination with
lake morphology and chemistry, determine the resultant
fish community (Smith and Powell 1971; Tonn 1990;
Poff 1997; Hershey et al. 1999; Jackson et al. 2001).
As an aside, we advocate caution when interpreting
community relationships with lake isolation because a
lake’s relative position in the landscape can also influence lake morphometry (i.e., size and shape) and a variety of limnological characteristics, such as the concentration of major ions and the vertical distribution of primary production (D’Arcy and Carignan 1997; Kratz et
al. 1997; Riera et al. 2000). The lack of correlation between all measures of lake isolation and lake morpholo-
582
gy and water chemistry suggests that our results are not
confounded by this interaction, and thus we are more
confident in interpreting relationships between lake isolation and fish community structure among our study
lakes.
Lake patterns in fish communities, lake isolation
and environmental factors
Our results show that the spatial arrangement of the
lakes, characteristics of the interconnecting streams and
lake morphology correspond strongly to patterns in fish
community composition. However, at the scale of individual lakes, local fish communities were found to exhibit varying levels of correlation with lake isolation and
environmental conditions. In many cases, our results
showed that patterns in community structure were better
explained by isolation-related factors compared to lake
morphology and water chemistry. This supports the hypothesis that the insular nature of many lakes could mask
associations between fish community composition and
the lake habitat conditions. Angermeier and Schlosser
(1989) suggested that when extinction and colonization
processes are important, there should be a lack of relationship between assemblage structure and habitat characteristics. Snodgrass et al. (1996) employed the same
argument when they suggested that the lack of relationship between wetland size and community richness in
isolated wetlands of the Upper Coastal Plain (USA)
could result from the dependence between fish community structure and differences in immigration rates between species. Our results support this perspective. For
example, Big Orillia Lake (lake 14), Little Orillia Lake
(lake 15) and Teapot Lake (lake 16) contain unusual fish
communities for their small area and shallow size, including the presence of two cool- or cold-water species,
brook trout and cisco. Consequently, it is not surprising
that these lakes exhibited poor agreement with expectations based on lake morphology and water chemistry;
however, they were shown to have high concordance
with three isolation measures: HG-tree; watercourse distance among lakes excluding intermediate water bodies
(WC-body); and stream gradient (D-elev1). Although
environment conditions may not be optimal for such species in these communities, the fact that the lakes are well
connected to other water bodies (in terms of WC-dis and
elevation) facilitates the persistence of the species. In
this situation, these lakes possibly act primarily as recipients of immigrating species (i.e., sinks) from adjacent
systems (i.e., sources). The notion of source-sink populations in aquatic communities has been suggested for coral reef fishes (Sale 1978, 1979) and stream fishes
(Osborne and Wiley 1992; Schlosser 1995; Fraser et al.
1999), with each of these studies providing evidence
supporting the importance of immigration from source
areas for shaping species richness and fish community
structure in potential sink habitats. We believe that
source-sink dynamics or true metapopulations can exist
in these study lakes (and drainage lakes in general),
which perhaps exhibit sub-optimal biotic and environmental conditions but are highly connected to other lakes
with suitable conditions. However, to establish whether
such lakes contain true metapopulations (Pulliam 1988;
Watkinson and Sutherland 1995) would require detailed
measurements of birth, death, immigration and emigration rates, which would be challenging to assess with
fish communities. Nevertheless, our results indicate candidate lakes where the study of such relationships may
be informative.
Similar to the above example, the fish community
composition of Sunken Lake (lake 34) illustrated poor
agreement based on its morphological characteristics,
but showed good agreement based on its degree of isolation and water chemistry. Sunken Lake has an unusual
fish assemblage, containing a depauperate community
with the presence of only yellow perch, brown bullhead
and smallmouth bass. The lack of smaller-bodied species
(e.g., cyprinids) in this lake is noticeable, since its habitat conditions appear to be suitable for another community given the large residuals between habitat and fish
community ordinations. The lack of agreement between
community composition and habitat may arise from
strong predation pressure by smallmouth bass and/or due
to the steep stream gradient restricting or even blocking
the immigration of small-bodied species. Previously existing populations of small-bodied fishes could have
been eliminated when bass colonized the lake (likely as a
result of unauthorized stocking), and have not been “rescued” by re-colonization from surrounding water bodies
(Ray and Gilpin 1991). Interestingly, a possible source
population [i.e., Raven Lake (lake 31)] contains 13 species (a large number of which are cyprinids) and is located immediate downstream from Sunken Lake (3.2 km
and 16 m in elevational difference). This example provides a good illustration of the potential inter-play between insular and habitat-related forces shaping the composition of fish communities. A number of experimental
studies have examined interactions between isolation and
habitat suitability (e.g., Meffe and Sheldon 1990; Peterson and Bayley 1993; Sheldon and Meffe 1994). Recently, Lonzarich et al. (1998) removed fish from pools in
two Arkansas streams and showed that four measures of
pool isolation (riffle length, riffle depth, position in
drainage basin and distance to large pools) were related
to the numerical recovery of the communities. For example, the recovery of an assemblage separated from neighbouring pools by short riffles occurred in 30 days whereas more isolated pools had not reached 70% recovery after 40 days. In addition, numerical recovery was more
rapid in downstream pools and in pools that were closer
to large source pools. Therefore, our study supports the
idea of Lonzarich et al. (1998) that although habitat
might be appropriate for the persistence of particular fish
species, the degree of habitat isolation is a major factor
structuring the composition of fish communities.
For a number of lakes, fish community composition
showed either good agreement with expectations based
583
on lake isolation and morphology (e.g., Jill Lake, Herb
Lake, Poorhouse Lake) or with expectations based on
lake isolation and water chemistry (e.g., Little Wren
Lake, Ernest Lake, Kawagama Lake), but not with both
lake morphology and water chemistry. Given the observed and extensive movements of tagged fish between
many of the study lakes (Jackson, unpublished data), we
believe that colonization and extinction rates are interconnected in such cases. This makes intuitive sense
because both isolation and habitat-related factors will influence the potential for species re-establishment following an environmental or ecological perturbation, and
therefore influence the probability of local extinction
(Magnuson et al. 1998). For example, the influx of individuals from adjacent lakes could provide a rescue effect
for populations containing few individuals. This may
prevent numbers from dropping so low as to lead to local
extinction. In situations where movements are very frequent, a local population may be lost due to extreme environmental fluctuations (e.g., winterkill due to low dissolved oxygen) but individuals could re-colonize before
periodic sampling detects the loss of the species.
As the crow flies or as the fish swims?
Frequently used isolation measures such as the straightline distance between lakes provides a simple measure of
the relative location of the lakes in the watershed. However, this measure will be misleading in convoluted systems where lakes close to each other in the straight-line
sense are quite far apart as measured by in-water distance (e.g., Little et al. 1997). Consequently, straight-line
distance will often not provide insight regarding the biologically relevant distances required for dispersal,
whereas the other measures we present do. The inadequacy of straight-line distance is not only conceptually
obvious, but it is also evident for a number of the study
lakes. For instance, the fish communities of Blue Chalk
Lake (lake 23) and Wren Lake (lake 28) showed good
agreement with all measures of lake isolation except
SL-dis. In this case, using such a simple measure of lake
isolation would be quite misleading since one might conclude that local patterns in community composition are
not influenced by the lakes’ degree of isolation from other lakes in the watershed. Another example includes
South McDonald Lake (lake 13) and Teapot Lake (lake
16) whose fish communities show strong concordance
with expectations based on D-elev1, and WC-body and
HG-tree, respectively, but both show weak concordance
with all other measures of isolation, lake morphology
and water chemistry. Again, measures incorporating other factors, such as the WC-dis and changes in elevation
enhance the detection of isolation processes shaping species composition. Finally, an interesting finding is that
the fish assemblage of Big East Lake (lake 3) showed a
stronger association with D-elev2, a measure describing
the relative magnitude of the stream elevational gradient
between the lakes, but do not match with D-elev1, a
measure of the absolute WC-dis and change in elevation
between the lakes. Apparently, the total differences in
WC-dis and elevation to surrounding lakes are not important isolating factors for this lake; however, in combination as a measure of the elevational gradient, the relationship between these two factors represents a potential
isolating mechanism. Since each isolation measure explained different local patterns in fish assemblage structure, we advocate the examination of various alternate
measures of lake interconnectedness. This is advantageous because different aspects of the landscape can influence the ability of species to disperse, and ultimately
can provide insight into determinants influencing community similarity among lakes. Optimally, alternative
measures should describe watercourse- and elevation-related aspects of the landscape separating the lakes, such
as US-sym, WC-dis and D-elev2.
Conclusion
Both isolation and environment-related factors can be important and should be examined for their effects in structuring fish communities. Identifying the degree of spatial
isolation among lakes is important for distinguishing biogeographical factors that are potentially responsible for
differences in species composition across watersheds. Our
paper outlined various means of calculating the spatial
connectivity between lakes, emphasizing different mechanisms that could potentially promote or prohibit the movement of fish. Inter-lake distances defined “as the fish
swims” provide greater insight in species-isolation relationships compared to distances defined “as the crow
flies”. Although patch or metapopulation dynamics of fish
assemblages have not been studied to any great extent, the
potential role of fish movement in contributing to homogeneity or heterogeneity in species community composition must be considered. Isolation measures incorporating
the difficulties fish encounter when traveling between water bodies will aid in identifying potentially insular drainage lakes within a watershed, as well as aid in understanding the role of isolation-related factors in shaping the
composition of fish communities. An objective of our
study was not to advocate the use of one “best” isolation
measure for fish community studies as this may vary depending on the data set considered (e.g., relative variation
in distances and elevational changes among lakes). Rather,
we hope to illustrate that alternative methodologies for
quantifying lake isolation should be examined when assessing the influence of spatial isolation on the structuring
of fish communities. Furthermore, depending on the topographic characteristics of the landscape, researchers could
use a subset of these lake isolation measures emphasizing
different attributes in their data. For instance, for lakes located in a landscape of intermediate or high topographic
relief, measures incorporating stream slope (i.e., D-elev1,
D-elev2) and upstream distance (i.e., US-sym, US-skew)
would be recommended, whereas in highly convoluted
systems having limited elevational differences, water-
584
course distance (i.e., WC-dis, WC-body) could be optimal. Finally, the Procrustean methodology presented here
provides a means of assessing the degree of overall concordance between biological and spatial multivariate data
matrices as well as providing a more detailed interpretation of the relative concordance of different data sets for
individual observations. This statistical methodology is a
promising quantitative tool not only for aquatic studies,
but also terrestrial studies. In the present study, Procrustes
analysis provided a robust assessment of the importance
of isolation- and habitat-related factors at the landscape
scale, and also facilitated the identification of factors influencing local fish community composition. Recognizing
that controlling mechanisms can vary within a spatial
framework may help ecologists study the way in which
physical and biological processes interact in the regulation
of lake fish communities.
Acknowledgements We would like to thank John Magnuson,
Brian Shuter, Keith Somers, Bill Tonn and two anonymous reviewers for their comments on early drafts. Also, we thank Pierre
Legendre for his helpful remarks regarding skew-symmetric matrices. Funding for this project was provided by a Natural Sciences
and Engineering Council of Canada (NSERC) Graduate Scholarship and University of Toronto scholarships to J.D.O, an NSERC
Research Grant to D. A. J. and a Brazilian CNPq scholarship to P.
R. P.-N. PROTEST software is available at http://www.200.utoronto.ca/jackson/.
References
Angermeier PL, Schlosser IJ (1989) Species-area relationships for
stream fishes. Ecology 70:1450–1462
Barbour CD, Brown JH (1974) Fish species diversity in lakes. Am
Nat 108:473–489
Brown JH (1971) Species richness of boreal mammals living on
the montane mountains of the Great Basin. Am Nat 105:
467–478
Chapleau F, Findlay CS, Szenasy E (1997) Impact of piscivorous
fish introductions on fish species richness of small lakes in
Gatineau Park, Quebec. Ecoscience 4:259–268
Chen Y, Jackson DA, Paloheimo JE (1994) Robust regression approach to analyzing fisheries data. Can J Fish Aquat Sci
51:1420–1429
D’Arcy P, Carignan R (1997) Influence of catchment topography
on water chemistry in southeastern Quebec Shield lakes. Can J
Fish Aquat Sci 54:2215–2227
Digby PNG, Kempton RA (1987) Multivariate analysis of ecological communities. Chapman and Hall, New York
Eadie JM, Keast A (1984) Resource heterogeneity and fish species
diversity in lakes. Can J Zool 62:1689–1695
Eadie JM, Hurly TA, Montgomerie RD, Teather KL (1986) Lakes
and rivers as islands: species-area relationships in the fish faunas of Ontario. Environ Biol Fishes 15:81–89
Fahrig L, Merriam G (1994) Conservation of fragmented populations. Conserv Biol 8:50–59
Findlay CS, Bert DG, Zheng L (2000) Effect of introduced piscivores on native minnow communities in Adirondack lakes.
Can J Fish Aquat Sci 57:570–580
Fraser DF, Gilliam JF, MacGowan MP, Arcaro CM, Guillozet PH
(1999) Habitat quality in a hostile river corridor. Ecology
80:597–607
Gauch HG (1982) Multivariate analysis in community ecology.
Cambridge University Press, Cambridge
Gower JC (1975) Generalized procrustes analysis. Psychometrika
40:33–51
Hanski I (1982) Dynamics of regional distribution: the core and
satellite species hypothesis. Oikos 38:210–221
Hanski I (1991) Metapopulation dynamics: brief history and conceptual domain. Biol J Linn Soc 42:3–16
Harrison S, Murphy DD, Ehlich PR (1988) Distribution of the Bay
checkerspot butterfly, Euphydryas editha bayensis: evidence
for a metapopulation model. Am Nat 132:360–382
Harvey HH (1975) Fish populations in a large group of acidstressed lakes. Int Verh Theor Angew Limnol 19:2406–2417
Harvey HH (1978) Fish communities of the Manitoulin Island
lakes. Int Verh Theor Angew Limnol 20:2031–2038
Harvey HH (1981) Fish communities of the lakes of the Bruce
Peninsula. Int Verh Theor Angew Limnol 21:1222–1230
Hershey AE, Gettel GM, McDonald ME, Miller MC, Mooers H,
O’Brien WJ, Pastor J, Richards C, Schuldt JA (1999) A geomorphic-trophic model for landscape control of arctic lake food
webs. Bioscience 49:887–897
Jackson DA (1988) Fish communities in lakes of the Black and
Hollow River watersheds, Ontario. MSc thesis. University of
Toronto, Toronto
Jackson DA (1995) PROTEST: a procrustean randomization test
of community environment concordance. Ecoscience 2:297–
303
Jackson DA, Harvey HH (1989) Biogeographic associations in
fish assemblages: local versus regional processes. Ecology
70:1472–1484
Jackson DA, Harvey HH (1993) Fish and benthic invertebrates
community concordance and community-environment relationships. Can J Fish Aquat Sci 50:2641–2651
Jackson DA, Harvey HH (1997) Qualitative and quantitative sampling of lake fish communities. Can J Fish Aquat Sci 54:
2807–2813
Jackson DA, Somers KM (1989) Are the probability estimates
from the permutation model of Mantel’s test stable? Can J
Zool 67:766–769
Jackson DA, Somers KM, Harvey HH (1992) Null models and
fish communities: evidence of nonrandom patterns. Am Nat
139:930–951
Jackson DA, Peres-Neto PR, Olden JD (2001) What controls who
is where in freshwater fish communities – the roles of biotic,
abiotic and spatial factors. Can J Fish Aquat Sci 58:157–170
Johnson MG, Leach JH, Minns CK, Olver CH (1977) Limnological characteristics of Ontario lakes in relation to associations
of walleye Stizostedion vitreum vitreum, northern pike Esox
lucius, lake trout Salvelimus namaycush, and smallmouth bass
Micropterus dolomieui. J Fish Res B Can 34:1592–1601
Karr JR (1981) Assessment of biotic integrity using fish communities. Fisheries (Bethesda) 6:21–27
Kratz TK, Webster KE, Bowser CJ, Magnuson JJ, Benson BJ
(1997) The influence of landscape position on lakes in northern Wisconsin. Freshwater Biol 37:209–217
Legendre P, Legendre L (1998) Numerical Ecology. Developments
in environmental modelling, vol 20, 2nd edn. Elsevier, Amsterdam
Levins R (1969) Some genetic and demographic consequences of
environmental heterogeneity for biological control. Bull Entomol Soc USA 15:237–240
Little LS, Edwards D, Porter DE (1997) Kriging in estuaries: as
the crow flies, or as the fish swims? J Exp Mar Biol Ecol
213:1–11
Lonzarich, DG, Warren, ML, Lonzarich ME (1998) Effects of
habitat isolation on the recovery of fish assemblages in experimentally defaunated stream pools in Arkansas. Can J Fish
Aquat Sci 55:2141–2149
MacArthur RH, Wilson EO (1967) The theory of island biogeography. Princeton University Press, Princeton, N.J.
MacRae PSD, Jackson DA (2001) The influence of smallmouth
bass (Micropterus dolomier) predation and habitat complexity
on the structure of littoral zone fish assemblages. Can J Fish
Aquat Sci 58:342–351
Magnan P, Rodriguez MA, Legendre P, Lacasse S (1994) Dietary
variation in a freshwater fish species: relative contributions of
585
biotic interactions, abiotic factors, and spatial structure. Can J
Fish Aquat Sci 51:2856–2865
Magnuson JJ (1976) Managing with exotics – a game of chance.
Trans Am Fish Soc 105:1–9
Magnuson JJ, Tonn WM, Banerjee A, Toivonen J, Sanchez O,
Rask M (1998) Isolation vs. extinction in the assembly of fishes in small northern lakes. Ecology 79:2941–2956
Matthews, WJ, Robison HW (1998) Influence of drainage connectivity, drainage area and regional species richness on fishes of
the interior highlands in Arkansas. Am Midl Nat 139:1–19
Meffe GK, Sheldon AL (1990) Post-defaunation recovery of fish
assemblages in southeastern blackwater streams. Ecology 71:
657–667
Osborne LL, Wiley MJ (1992) Influence of tributary spatial position on the structure of warmwater fish communities. Can J
Fish Aquat Sci 49:671–681
Paine RT (1988) Habitat suitability and local population persistence of the sea palm Postelsia palmaeformis. Ecology 69:
1787–1794
Peterson JT, Bayley PB (1993) Colonization rates of fishes in experimentally defaunated warmwater streams. Trans Am Fish
Soc 122:199–207
Poff NL (1997) Landscape filters and species traits: towards
mechanistic understanding and prediction in stream ecology. J
N Am Benthol Soc 16:391–409
Pulliam HR (1998) Sources, sinks, and population regulation. Am
Nat 132:652–661
Radomski PJ, Goeman TJ (1995) The homogenizing of Minnesota
fish assemblages. Fisheries (Bethesda) 20:20–23
Rahel FJ (1986) Biogeographical influences on fish species composition of northern Wisconsin lakes with applications for lake
acidification studies. Can J Fish Aquat Sci 43:124–134
Ray C, Gilpin M (1991) The effect of conspecific attraction on
metapopulation dynamics. Biol J Linn Soc 42:123–134
Riera JL, Magnuson JJ, Kratz TK, Webster KE (2000) A geomorphic template for the analysis of lake districts applied to
Northern Highland Lake District Wisconsin, USA. Freshwater
Biol 43:301–318
Robinson CLK, Tonn WM (1989) Influence of environmental factors and piscivory in structuring fish assemblages of small Alberta lakes. Can J Fish Aquat Sci 46:81–89
Rohlf FJ (1990) Rotational fit (Procrustes) methods. In: Rohlf FJ,
Bookstein FL (eds) Proceedings in the Michigan Morphometrics Workshop. Museum of Zoology special publication no. 2.
University of Michigan, Ann Arbor, Mich., pp 227–236
Rohlf FJ, Slice D (1990) Extensions of the procrustes method for
the optimal superimposition of landmarks. Syst Zool 39:40–59
Sale PF (1978) Coexistence of coral reef fishes – a lottery for living space. Environ Biol Fish 3:85–102
Sale PF (1979) Recruitment, loss and coexistence in a guild of territorial coral reef fishes. Oecologia 42:159–177
Saunders DA, Hobbs RJ, Margules CR (1991) Biological consequences of ecosystem fragmentation: a review. Conserv Biol
5:18–32
Schlosser IJ (1995) Dispersal, boundary processes, and trophic
level interactions in streams adjacent to beaver ponds. Ecology
76:908–925
Sheldon AL, Meffe, GK (1994). Short-term recolonization by
fishes of experimentally defaunated pools of a coastal plain
stream. Copeia 1994(4):828–837
Siegel AF (1982) Robust regression using repeated medians.
Biometrika 69:242–244
Siegel AF, Benson RH (1982) A robust comparison of biological
shapes. Biometrics 38:341–350
Sjögren Gulve P (1994) Distribution and extinction patterns within a northern metapopulation of the pool frog, Rana lessonae.
Ecology 75:1357–1367
Slice DE (1994) GRF-ND: generalized rotational fit on n-dimensional landmark data. Department of Ecology and Evolution,
State University of New York at Stony Brook, Stony Brook,
N.Y.
Smith CL, Powell CR (1971) The summer fish communities of
Brier Creek, Marshall County, Oklahoma. Am Museum Novitates 2458:1–30
Snodgrass JW, Meffe GK (1998) Influence of beavers on stream
fish assemblages: effects of pond age and watershed position.
Ecology 79:928–942
Snodgrass JW, Bryan AL, Lide RF, Smith GM (1996) Factors affecting the occurrence and structure of fish assemblages in isolated wetlands of the Upper Coastal Plain, U.S.A. Can J Fish
Aquat Sci 53:443–454
Tonn WM (1990) Climate change and fish communities: a conceptual framework. Trans Am Fish Soc 119:337–352
Tonn WM, Magnuson JJ (1982) Patterns in the species composition and richness of fish assemblages in northern Wisconsin
Lakes. Ecology 63:1149–1166
Tonn WM, Magnuson JJ, Rask M, Toivonen J (1990) Intercontinental comparison of small-lake fish assemblages: the balance
between local and regional processes. Am Nat 136: 345–375
Townsend CR, Crowl TA (1991) Fragmented population structure
in a native New Zealand fish: an effect of introduced brown
trout? Oikos 61:347–354
Watkinson AR, Sutherland WJ (1995) Sources, sinks and pseudosinks. J Anim Ecol 64:126–130